Cost-effectiveness analysis of malaria rapid diagnostic test incentive schemes for informal private healthcare providers in Myanmar

Background The emergence of artemisinin-resistant Plasmodium falciparum parasites in Southeast Asia threatens global malaria control efforts. One strategy to counter this problem is a subsidy of malaria rapid diagnostic tests (RDTs) and artemisinin-based combination therapy (ACT) within the informal private sector, where the majority of malaria care in Myanmar is provided. A study in Myanmar evaluated the effectiveness of financial incentives vs information, education and counselling (IEC) in driving the proper use of subsidized malaria RDTs among informal private providers. This cost-effectiveness analysis compares intervention options. Methods A decision tree was constructed in a spreadsheet to estimate the incremental cost-effectiveness ratios (ICERs) among four strategies: no intervention, simple subsidy, subsidy with financial incentives, and subsidy with IEC. Model inputs included programmatic costs (in dollars), malaria epidemiology and observed study outcomes. Data sources included expenditure records, study data and scientific literature. Model outcomes included the proportion of properly and improperly treated individuals with and without P. falciparum malaria, and associated disability-adjusted life years (DALYs). Results are reported as ICERs in US dollars per DALY averted. One-way sensitivity analysis assessed how outcomes depend on uncertainty in inputs. Results ICERs from the least to most expensive intervention are: $1,169/DALY averted for simple subsidy vs no intervention, $185/DALY averted for subsidy with financial incentives vs simple subsidy, and $200/DALY averted for a subsidy with IEC vs subsidy with financial incentives. Due to decreasing ICERs, each strategy was also compared to no intervention. The subsidy with IEC was the most favourable, costing $639/DALY averted compared with no intervention. One-way sensitivity analysis shows that ICERs are most affected by programme costs, RDT uptake, treatment-seeking behaviour, and the prevalence and virulence of non-malarial fevers. In conclusion, private provider subsidies with IEC or a combination of IEC and financial incentives may be a good investment for malaria control. Electronic supplementary material The online version of this article (doi:10.1186/s12936-015-0569-7) contains supplementary material, which is available to authorized users.


Decision Tree Overview Concepts
We designed a decision tree using a spreadsheet in Microsoft Excel, 2010. A decision tree consists of decision nodes (denoted by squares) and chance nodes (denoted by circles). Each branch that arises from a chance branch is mutually exclusive of other branches, and the combined possibilities from each node must be collectively exhaustive such that branch probabilities add to 1. A decision tree that compares health service delivery options conventionally starts with each incentive method denoted as a decision option.
The decision tree is a flow chart that aims to encompass all possible courses of action within a chosen population (Figure 1). The population for this study comprises of febrile patients in six townships within Myanmar's Mon and Shan states that seek healthcare within the informal private sector from May to September. By following the decision tree from left to right, each pathway of action is represented by a terminal node (denoted '//'). The decision node represents subsidy schemes while chance nodes are categorized as: disease state, provider behavior, diagnosis, and health outcomes. In the beginning of the decision tree, each subsidy method is represented as a decision We use 'no intervention' as a base case for reference. For each subsidy scheme, we use a conventional decision tree structure for diagnostic tests, which starts with the true disease state as the first chance node. Our disease states are: P. falciparum malaria or more (including P. vivax), P. vivax malaria only, and no malaria, as we assume that all non-falciparum malaria is P. vivax malaria.
For each disease state, we then use a chance node for whether the provider uses an RDT. If the patient is tested with an RDT, there will be four possible test results for the combined pan Plasmodium test and P. falciparum diagnostic test: ++, +-, -+, and --. The probability of each result depends on whether the test result is true or false: a true positive corresponds to test sensitivity, false negative to (1 sensitivity), true negative to specificity, and false positive to (1 specificity). If the provider does not use an RDT, the resulting clinical diagnosis will be positive or negative for malaria. We assume that the pan Plasmodium test and P. falciparum test performance is independent: failure to detect malaria with the pan Plasmodium PLDH antigen-detecting test is uncorrelated with failure to detect malaria with the P. falciparum HRP2 antigen-detecting test.
Each test result then leads to one of three treatment possibilities that are chosen by the provider. The provider sells subsidized ACTs, other antimalarials, or no antimalarial to the patient. After each treatment, the patient either lives or dies, which is a terminal node represented by "//". If the patient lives, morbidity associated with illness is accounted for in quantified health outcomes.

Data Inputs
There are three types of data inputs described below. The actual sources of data are described in section 4 (Data Inputs).

Probability Data
A probability value is assigned to each node. Each decision node is assigned a probability of 1, and each chance node is assigned a total probability of 1. Since chance nodes encompass a mutually exclusive and collectively exhaustive set of possible actions, the sum of probabilities that emerge from each chance node adds to 1.
The probability of occurrence for each path leading to a terminal node is calculated as the product of probabilities of every chance node from the decision node to that terminal node. The probability of occurrence is used to weight the associated costs (measured in USD) and health outcomes (measured in DALYs) of each path.

Cost Data
Cost data is incorporated at every applicable node and measured in USD. The exchange rate used is 907 Kyats / dollar, the official rate on May 1 st 2013. Exchange rates in the model can be changed and updated easily. The total cost associated with each path is weighted by probability, and the sum of weighted costs for each decision node is the total cost for that decision node.

Health Outcome Data
The health outcome that corresponds to every terminal node is quantified in DALYs. Life years lost to mortality are calculated by subtracting the mean life expectancy from the average age of malaria-induced death and applying a 3% discount rate to each year in the future. Discounting is typically used in CEA to account for the fact that people tend to value events in the present more than in the distant future. For surviving patients, the morbidity of those who do not recover immediately will be calculated by dividing the average duration of illness by the DALY weight for the illness. The health outcomes associated with each path are weighted by probability, and their sum for each decision node represent the total DALYs incurred for that decision node.

Comparison of Intervention Arms
Once probabilities, weighted costs, and weighted health outcomes are established, the sum of weighted costs and health outcomes for each decision node (subsidy scheme) are used to compare the three intervention arms and the base case (no intervention). We compare the subsidy methods by ordering the approaches from least to most expensive. The subsidy methods were first obtained incrementally. The incremental cost-effectiveness ratio (ICER) for each intervention is obtained by the following formula: ICER = (Cost of interventioncost of next less expensive approach) (DALYs with intervention) -(DALYs with next less expensive approach)

Extended dominance
In this study, we found that increasing programmatic interventions led to increases in costs that were compensated by larger increases in effects. This led to decreasing ICERs. This is called "extended dominance", where a combination of 'no intervention' and the last intervention (arm 3) is more cost-effective than the intermediate interventions (arm 1). The proper comparison is thus each intervention versus the reference, not versus the next-least expensive option. The formula used in this case: Cost per DALY averted vs baseline = ____(Cost of interventioncost of no intervention) (DALYs with intervention) -(DALYs from no intervention)

Patients adhere to a full 3-day course of subsidized ACTs.
Rationale: the AMTR project uses a multipronged approach to encourage the completion of three-day ACT regimens: 1) the price of a full course of ACTs is set to match the price of partial courses of artemisinin monotherapy that patients afford 2) both provider support visits and community outreach programs emphasize the importance of completing a full course of ACTs 3) the design of the ACT packaging includes two written Burmese reminders to complete a full course of ACTs: one on the front of the cardboard envelope and a second below the pills inside the envelope. The RDT household surveys will confirm whether this assumption is accurate: otherwise the model will be updated accordingly.

'Other antimalarial' refers to the use of quinine or chloroquine.
Rationale: The 2012 AMTR mystery client survey showed that only 8% of fever diagnoses were treated with non-artemisinin antimalarials. We chose quinine or chloroquine based on in-depth interview stock audit data from the pilot study, triangulated with conversation with PSI Myanmar program staff. Interestingly, none of the providers screened carried primaquine, which is the only drug combination capable of clearing hypnozoites, the latent liver stage of P. vivax infections.

Drug adherence to chloroquine and injectable quinine is high given the short course of therapy.
Rationale: Consultation with PSI Myanmar program officers suggested that patients seeking private sector care typically adhere to the first 2-3 days of a drug regimen. Pilot study qualitative demographic data shows that most providers who carry quinine carry it as an injectable solution. The injection is available as a single dose, and orally administered chloroquine is available as a three-day course of therapy.

P. vivax malaria does not relapse.
Rationale: The complexity of relapse and unavailability of epidemiological data prevent the accurate prediction of P. vivax relapse. A full course of primaquine is required to ensure the clearance of hypnozoites, the parasite stage responsible for the relapse of P. vivax malaria. Relapse rates depend on the duration of fever before initial treatment, the type of treatment used, the level of parasitemia, and the level of patient drug adherence. 4 Relapse rates are also likely to be low within the 1-month time frame considered by the model: a study at the Thai-Myanmar border showed a 28-day relapse rate of 3.4% for self-administered therapy and 0% for directly-observed and primaquine therapy. 4

'No antimalarial' refers to the use of antipyretics 70% of the time and antibiotics 30% of the time.
Rationale: A 2012 mystery client survey at PSI Myanmar showed that when antimalarials were not prescribed for fevers, 50% of cases were treated with antibiotics and the other 50% with antipyretics. However, more recent household surveys at PSI Myanmar showed that the vast majority (90% estimated) of providers administered antipyretics to patients presenting with a fever. We therefore assume that patients receive antipyretics 70% of the time.

Only one type of medication is prescribed at any given time.
Rationale: While some providers in Myanmar are known to administer "machine gun therapy" by prescribing multiple drugs, the 2012 PSI mystery client survey shows that only 0.4% of providers administered more than one drug at a given time.

Intervention details
The pilot study took place from May to September, 2013. The 631 outlets enrolled were in the RDT pilot study ( Table 1). The pilot study townships are shown in the map in Figure 2, of which Artemisinin Monotherapy Replacement (AMTR) outlets targeted for RDT scaleup are shown in red.

Input data sources
A combination of finance/account records and management information systems (MIS) data from PSI Myanmar, mixed methods data from the RDT pilot study (household surveys, interviews with private providers, mystery client visits, stock audit data from supply points), and a review of published scientific literature were used as detailed below.

Study Population and Epidemiology
The study population was defined as the number of patients that seek private provider care at all outlets enrolled in the RDT pilot study from May to September, 2013. Study population data was derived from malaria epidemiology and provider assessments from other PSI Myanmar programs and a review of published scientific literature.

Malaria Epidemiology and Provider Assessments from PSI Myanmar
PSI Myanmar has detailed Management Information System (MIS) data on malaria RDT uptake as well as provider-reported RDT test results and patient demographics. The data is a part of the PSI Sun Primary Health (SPH) social franchising brand for private providers throughout rural Myanmar. 5 We also used SPH monitoring mechanisms to estimate baseline provider knowledge levels of RDT use. A patient simulation known as the Sustained Quality study in 2011 used direct observation, clinical vignettes, and medical mannequins to assess provider understandings of malaria diagnosis and treatment before and after a single RDT training session (time points: 6 and 12 months post training). The study assessed provider knowledge levels in the following categories: medical history taking, looking for signs of severe malaria, checking for vital signs, malaria drug history, RDT performance, proper referral to higher level facilities, and drug prescription and information.

RDT Pilot Study Data
The RDT pilot study used four analytical methods: 1) household surveys, 2) interviews with private providers, 3) mystery client visits, and 4) stock audit data from supply points. Each method was used to evaluate RDT uptake and intervention effectiveness. The household surveys measured RDT uptake at a community / patient level before and after the intervention. The interviews with private providers explored provider attitudes to RDT use at the end of the pilot study and also included a survey of antimalarial and RDT stock and prices. Mystery clients presented with an alleged fever to providers that are enrolled in the pilot study as an additional measure for RDT uptake in the beginning and end of the pilot study. Stock audit data monitored RDT and ACT use, which was possible through the program requirement that used RDTs be returned to supply points for resupply. Providers were instructed to record the date, result of the test, and patient age / gender on each test. A comparison of RDT results with ACT sales suggested whether ACTs are being sold as recommended (for P. falciparum malaria only).

Cost Data from PSI Myanmar
Cost data was derived from PSI Myanmar finance and account records, and the exchange rate used is 907 Kyats / dollar, the official rate on May 1 st 2013. 6 Cost-effectiveness analysis from a societal perspective included operational costs to PSI Myanmar, commodity costs across the supply chain, as well as time and commodity costs to the patient and provider.
Operational costs were informed by PSI Myanmar finance and account data. To allow for the prediction of long-term operational costs, we defined operational costs in two categories: the cost of program initiation and the cost of recurrent program activities.
Program initiation consisted of staff training sessions, provider recruitment activities, and community education sessions on the utility and availability of subsidized RDTs. Community education sessions were considered to be a program initiation cost because once all intervention areas were reached, community education was considered complete and sessions were no longer continued. PSI program staff members expect that for scale-up, six months of community education sessions are sufficient to reach communities throughout the intervention areas. Staff training sessions are also a program initiation cost: PSI training program designers expect the initial staff training session to be much more involved and costly than additional refresher training sessions which take place every 6-12 months. Refresher training costs were not included in this analysis. The costs of research and preparation required for program initiation were also excluded.
Recurrent program activities included all field activities and office support required for program rollout. Field activities included the management and delivery of provider visits and community education sessions, and office support included administrative and managerial staff members. We excluded the cost of pilot study evaluative methods (i.e. mystery client visits) since research evaluative methods do not represent scale-up practices.
For both program initiation and recurrent program activities, the RDT pilot study had three types of operational costs: 1) overhead costs for PSI Myanmar, 2) staff costs, and 3) program materials and supplies. All costs were scaled to estimate the monthly operational costs for each arm of the RDT pilot study.
Commodity costs were mapped across the supply chain, starting from manufacturing and ending at the patient's drug or RDT purchase ( Figure 2). Costs included product costs and delivery costs. RDT costs from manufacturing to pharmacy supply point delivery were recurrent operational costs. RDT purchases from pharmacy supply points to private outlets are provider costs, and individual RDT purchases from private informal outlets are patient costs.

Figure 2. Malaria RDT supply chain for RDT pilot study
Patient and provider commodity costs were estimated through data from PSI programmatic activities as well as staff consultation. There were two sources of baseline data that are available from the AMTR program evaluation in late 2012: a full stock audit and a mystery client survey. The RDT program is nested within the AMTR program, which has been delivering subsidized ACTs to 6,865 private outlets throughout eastern Myanmar since September 2012. The AMTR stock audit reached 1,159 private outlets that were enrolled to purchase subsidized ACTs, recording all available antimalarials and RDTs. The mystery client survey reached 446 private providers to establish baseline provider behavior for patients that present with fever. The mystery client survey also recorded antimalarial and RDT stock and prices, the proportion of providers that used RDTs, the drugs recommended for sale to the patient, and the level of provider instruction to the patient on ACT use. Patient and provider time costs were estimated through informal discussion with PSI Myanmar staff members. Patient traveling costs are estimated from a similar CEA in scientific literature. We excluded the potential lost income from healthcare seeking and illness due to the unavailability of data.

Health Outcome Data
Health outcomes were quantified in Disability Adjusted Life Years (DALYs), with a 3% discount rate applied.  (Table 6), therefore the estimated number of clients is between 1/16% and 4/9%: 6 to 44.

Data Input values and rationale
* MIS data is from SPH interventions from July to October 2012 in the same Mon state townships as the RDT study. The sample includes 3769 patients that were tested for malaria within 24 hours of the onset of fever. MIS data from the Shan state was not available. ** Returned RDTs were both read by providers (the results were recorded on the RDT using a black permanent marker) as well as PSI staff. The reads between provider reports and PSI staff showed high concordance, and we chose to use provider reports since the rate of false positive RDT test results increases past the recommended 20-minute readout. We estimate that 2/3*30% (= 20%) of nonmalarial fevers get treated properly, the remaining 80% suffer the same fatality rate as those given ACT or other antimalarial.     * Numbers adjusted to be the same as no intervention due to small sample sizes ** 70% use of ACTs is an average value between the Mon and Shan state results from the 2012 mystery client survey. We scale the mystery client prescription of other antimalarials and no antimalarial (antibiotic and antipyretic) to a 70% ACT uptake. *** Numbers adjusted to match arm 1 since no data available for arms 2 and 3 **** Lowered estimate of actual ACT use because the 2012 mystery client survey took place 4 months after ACTs were first introduced, possibly leading to temporary overuse.

RDT Costs
RDT commodity costs include the cost of subsidy and are separated as costs to the donor, provider and patient ( Table 6). The RDT commodities used are: Standard Diagnostics (SD) Bioline Ag Pf Pv from Korea (used in the first 2 months of the intervention) and First Response, Premier Medical Corporation from India (used in the last 4 months of the intervention). The cost of RDT distribution across the supply chain is detailed in Table 7. Calculations for RDT costs are based on product markup costs between distribution stages, estimated to be: 3% from wholesale to retail, 20% from retail to distributor, and 100% from vendor to patient based on discussion with PSI program staff managers (Table 8). Estimates are indicated by red text while hard numbers are denoted by black text.    Table 9 summarizes drug costs: when the cost of marketed drugs could not be obtained, we use wholesale drug prices to estimate the cost to the patient. Product markup costs between distribution stages are estimated to be: 3% from wholesale to retail, 20% from retail to distributor, and 67% from vendor to patient based on estimates from PSI program staff managers. Unsubsidized drugs are considered to be patient costs while the donor cost of subsidized ACTs is also calculated ( Table 10). Subsidized ACTs are sold as four products: Supa Arte 1, 2, 3 and 4. The Supa Arte products, marketed from Ipca laboratories limited in India, contain various doses of artemether and lumefantrine targeting different age ranges. We scaled the percentage of use for each Supa Arte product according to the percentage of individuals within each age range that were tested for malaria in the Mon state from July to October, 2012 to obtain a weighted average cost of an ACT (Table 10). The exchange rate used is the rate at the time of purchase: 845.94 Kyat/dollar. To obtain societal drug costs, consumer or unit costs were obtained where available. Markup rates throughout distribution were used to estimate societal costs (Table 11). Wholesale costs include the cost of commodity, packaging materials, and packaging labor. All costs up to the retail level are donor costs, while costs to vendors and customers are at the provider and patient costs, respectively.

Results beyond first year of intervention
Results presented in the main text of the paper focus on the first year of the intervention. For recurrent years, costs are lower as staff will not need new training, community outreach will not be performed, and less providers will have to be recruited. Costs for the first year and subsequent years is shown in the tables below.